Forecasting choice

The chapter points out the need to have a credible behavioural theory on which forecasting can be based, rather than correlations observed in the data. It then goes on to present methods to deal with a range of issues that arise in forecasting with choice models. It discusses procedures for validation and the choice among competing models. In considering aggregate and disaggregate approaches to forecasting, sources of error are relevant; forecasting using simulation introduces sampling noise, but can offer practical advantages, depending on the context. The technique of ‘pivoting’ can be helpful in improving accuracy. The problems of forecasting with discrete-continuous models are discussed. When forecasting over an extended period, the stability of the model must be considered, as must the need to forecast the model inputs. New alternatives present a difficult problem. The chapter concludes with a discussion of model accuracy and how to assess it.